OpenAI's Deep Research Is Mid

In two words: overpriced prototype.

Im so glaď im not the only one reacting like that whenever agents are mentioned

Still unusable for edge cases and pretty much everything that involves long term reasoning and precision.

I've been using deep research for the past few days, and you're broadly spot on. The value you get out of it is based on how many open-ended research topics you want to ask it about. Inspired by a Veritasium video, I asked it for a survey of different formulations of the action in physics, and it gave good pointers that helped me much better understand the subject. I also asked it to survey deck-based automation mechanics in tabletop games, and, while it didn't mention everything, it was still very good. I also asked it to make a guide for turning educational material into narratives, and it did a pretty good job of listing both procedures to follow and what pitfalls to avoid. Of course, you need to exercise discretion, because DR won't; in that last report, it mentioned learning styles in passing uncritically even though it's pseudo science, for instance. But the myriad insights it gave beyond that made me quite satisfied. It's not a glorified replacement for a search engine; if you try to use it that way, you'll have a rough time. As long as you're thinking critically while reading its reports, you'll get a lot out of them.

QA testers used to get payed not pay themselves

9:49 "Gouge it's actual performance" 🧐🤨🤔

On the one hand I feel like deep research is just a knee jerk reaction to deep seek coming out with its reasoning stuff but at the same time I'm impressed if it works at all considering how fast they had to blitz it out

Deep research is like GPT-1, wait till the fourth/fifth generation releases ¯\_(ツ)_/¯

Check out Monica AI now, the best AI browsing companion! http://bit.ly/3Qh6zxy and use the code BYCLOUD10 for a special 10% off PS It is a really nitpicky video, a lot of capabilities is really limited to the model's performance, and not the information retrieval process. Currently on all other models that I tested, no model can properly explain DeepSeek's "cold-start" data yet. So I just wanna say again that Deep Research is amazing, especially with it capable of writing in long form. It's just that it might be on the model's part for a lot of its downfall, especially Deep Research's o3 is RL tuned specifically, which worsen its normal o3 capabilities.

You should be able to specify websites and files for the model to mandatory crawl into